[not loaded]
rdf_history.pl -- RDF Persistent store change history
This module deals with accessing the journal files of the RDF persistency layer to get insight in the provenance and history of the RDF database. It is designed for Wiki-like collaborative editing of an RDF graph. We make the following assumptions:
Users are identified using a URI, typically an OpenID (http://openid.net/)
Triples created by a user are added to a named graph identified by the
URI of the user.
Changes are grouped using rdf_transaction(Goal, log(Message, User))
The number that is associated with the named graph of a triple (normally
expressing the line number in the source) is used to store the time-stamp.
Although this information is redundant (the time stamp is the same as
for the transaction), it allows for binary search through the history
file for the enclosing transaction.
- rdfh_transaction(:Goal) is semidet
- Run Goal using rdf_transaction/2, using information from the HTTP layer to provide OpenID and session-id.
- rdfh_assert(+S, +P, +O) is det
- Assert a triple, adding current user and time to the triple context.
- rdfh_retractall(+S, +P, +O) is det
- Retract triples that match {S,P,O}. Note that all matching
triples are added to the journal, so we can undo the action as
well as report on retracted triples, even if multiple are
retracted at the same time.
One of the problems we are faced with is that a retract action goes into the journal of the user whose triple is retracted, which may or may not be the one who performed the action.
- rdfh_update(+S, +P, +O) is det
- More tricky stuff, replacing a triple by another. Typically this
will be changing the predicate or object. Provenance info should
move the new triple to the user making the change, surely if the
object is changed. If the predicate is changed to a related
predicate, this actually becomes less obvious.
Current simple-minded approach is to turn an update into a retract and assert. The S,P,O specifications are either a ground value or of the form Old
->New. Here is an example:rdfh_update(Work, Style, wn:oldstyle -> wn:newstyle)
- transaction_context(-Term) is det[private]
- Context to pass with an RDF transaction. Note that we pass the user. We don't need this for simple additions, but we do need it to track deletions.
- rdfh_session(-Session) is semidet[private]
- Session is a (ground) identifier for the current session.
- rdfh_user(-URI) is det[private]
- Get user-id of current session.
- rdfh_time(-Time:integer) is det[private]
- Get time stamp as integer. Second resolution is enough, and avoids rounding problems associated with floats.
- rdfh_triple_transaction(+Triple:rdf(S,P,O), -Transaction) is nondet
- True if the (partial) Triple is modified in Transaction.
- rdfh_db_transaction(?DB, +Condition, ?Transaction) is nondet
- True if Transaction satisfying Condition was executed on DB.
Condition is one of:
- true
- Always true, returns all transactions.
- id(Id)
- Specifies the identifier of the transaction. Only makes sense if DB is specified as transaction identifiers are local to each DB.
- after(Time)
- True if transaction is executed at or after Time.
- open_journal(+File, -Stream) is det[private]
- Open a journal file. Journal files are always UTF-8 encoded.
- journal_transaction(+JournalFile, ?Transaction) is nondet[private]
- True if Transaction is a transaction in JournalFile,
- seek_journal(+Fd:stream, +Spec) is semidet[private]
- See an open journal descriptor to the start of a transaction
specified by Spec. Spec is one of:
- after(Time)
- First transaction at or after Time. Fails if there are no transactions after time.
- id(Id)
- Start of transaction labeled with given Id. Fails if there is no transaction labeled Id.
The implementation relies on the incrementing identifier numbers and time-stamps.
- bsearch_journal(+Fd, +Start, +Here, +End, +Spec, !Last) is semidet[private]
- Perform a binary search in the journal opened as Fd.
- start_of_transaction(+Fd, +From, -Start, -Term) is semidet[private]
- Term is the start term of the first transaction after byte position From. Fails if no transaction can be found after From.
- rdfh_transaction_member(Action, Transaction) is nondet
- True if Action is an action in Transaction.